Migration 545a5b77e5e8 should handle missing NMA_Radionuclides.thing_id idempotently

[kbighorse]

Context

The current data model correctly uses Thing -> NMA_Chemistry_SampleInfo -> NMA_Radionuclides.

Problem

alembic/versions/545a5b77e5e8_add_chemistry_backfill_columns_to_.py upgrade drops NMA_Radionuclides.thing_id and its FK as if always present, while d9f1e2c3b4a5 already drops that column. This can fail in environments with different migration history/state.

Proposed fix

• Make 545a5b77e5e8 drop logic defensive (introspect columns/constraints before dropping).
• Keep downgrade restoration safe (add nullable, backfill from NMA_Chemistry_SampleInfo.thing_id, then enforce NOT NULL + FK).
• Add migration tests for both preconditions (thing_id present vs already absent).

Acceptance criteria

• alembic upgrade head succeeds in both preconditions.
• Downgrade of 545a5b77e5e8 succeeds with existing radionuclide rows.
• Runtime association model remains unchanged (NMA_Radionuclides linked via chemistry_sample_info_id).

[github-actions]

Linked Jira ticket created: BDMS-622